Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Modelling Scenarios for Carbon-aware Geographic Load Shifting of Compute Workloads

Wim Vanderbauwhede

TL;DR

This work quantifies the potential of geographic load shifting to reduce data-centre emissions by coupling embodied life-cycle carbon with operation-phase emissions in a simple two-site model. By applying the model to commercial AI data-centres and HPC scenarios, the authors show that even optimistic GLS configurations yield reductions around 4–5%, far from offsetting the projected growth in data-centre energy use. The study highlights critical factors—nonzero renewable carbon intensity, embodied emissions, idle power, and capacity/curtailment constraints—that cap the achievable gains. The results underscore the need for complementary decarbonisation strategies, including efficiency improvements, renewable integration, and lifecycle optimisations, to meaningfully curb ICT emissions at scale.

Abstract

We present an analytical model to evaluate the reductions in emissions resulting from geographic load shifting. This model is optimistic as it ignores issues of grid capacity, demand and curtailment. In other words, real-world reductions will be smaller than the estimates. However, even with these assumptions, the presented scenarios show that the realistic reductions from carbon-aware geographic load shifting are small, of the order of 5\%. This is not enough to compensate the growth in emissions from global data centre expansion.

Modelling Scenarios for Carbon-aware Geographic Load Shifting of Compute Workloads

TL;DR

This work quantifies the potential of geographic load shifting to reduce data-centre emissions by coupling embodied life-cycle carbon with operation-phase emissions in a simple two-site model. By applying the model to commercial AI data-centres and HPC scenarios, the authors show that even optimistic GLS configurations yield reductions around 4–5%, far from offsetting the projected growth in data-centre energy use. The study highlights critical factors—nonzero renewable carbon intensity, embodied emissions, idle power, and capacity/curtailment constraints—that cap the achievable gains. The results underscore the need for complementary decarbonisation strategies, including efficiency improvements, renewable integration, and lifecycle optimisations, to meaningfully curb ICT emissions at scale.

Abstract

We present an analytical model to evaluate the reductions in emissions resulting from geographic load shifting. This model is optimistic as it ignores issues of grid capacity, demand and curtailment. In other words, real-world reductions will be smaller than the estimates. However, even with these assumptions, the presented scenarios show that the realistic reductions from carbon-aware geographic load shifting are small, of the order of 5\%. This is not enough to compensate the growth in emissions from global data centre expansion.

Paper Structure

This paper contains 24 sections, 21 equations, 9 figures, 2 tables.

Table of Contents

  1. Context: Emissions from AI data centres
  2. Related work: Carbon-aware computing
  3. Research question: what is the potential of geographic load shifting
  4. Model for emissions reduction of geographic load shifting
  5. Model for emissions reduction of geographic load shifting
  6. Model parameters
  7. Embodied emissions
  8. Emissions from use
  9. Rationale for the model construction
  10. Baseline emissions model
  11. Emissions model including geographic load shifting
  12. Emission reductions from geographic load shifting
  13. Scenarios for geographical load shifting
  14. Scenarios for commercial data centres
  15. Scenario 1: Geographic load shifting to optimise use of solar energy
  16. ...and 9 more sections

Figures (9)

  • Figure 1: Years of growth as per the McKinsey scenarios compensated by emission reductions
  • Figure 2: Locations used in the scenarios. US, UK, Germany, France and Spain are used in the commercial data centre scenarios, the HPC sites (BNL, EPCC, HPC2N, ASCG) are used in the supercomputer scenarios.
  • Figure 3: Solar scenario, ideal assumptions (two sites are low-CI in any 8-hour period). 100 MW can be shifted.
  • Figure 4: Solar scenario, more realistic assumptions (two, three or one low-CI sites in an 8-hour period). Only (100+50+50)/3 MW can be shifted.
  • Figure 5: Wind scenario, ideal assumptions (wind load factor 50% for all sites, no correlation). 100 MW can be shifted.
  • ...and 4 more figures